J. Kissel, S. Dwyer, K. Kawabe
WP # 6448

Since we've been having trouble with the intermittent ring-up of the H1 SUS ETMY's 4735 Hz violin mode, Sheila and Keita suggest that we permanently turn on the 4735 Hz notch that's already in place in the LSC-DARM2 filter bank. Contrary to traditional belief, however, a change to the DARM filter banks no longer has no impact on the calibration, in general, because the time-dependent correction factors -- now applied to the h(t) data stream -- are computed using ratios of calibration lines, and the frequency dependence of full DARM loop shape (including changes to the DARM filter bank) between line frequencies must be compensated (see T1500377 and P1600063). 

In this particular case, one might expect a well-designed notch at 4.7 kHz should have negligible impact on calibration line frequencies used for time-dependence, given that the highest frequency line used is 331 Hz. However, the notch was not well-designed; just something slapped in quickly in a panic to prevent lock loss.

Thus, Sheila and I propose a new design. See attached. 
Called out explicitly here:
                  Current Design: notch(4735.25,5,80)
   Increased Q of Current Design: notch(4735.25,100,80)
             Proposed New Design: ellip("BandStop",4,1,80,4720,4750)gain(1.12202)

Recall that foton's notch functions take inputs of the form
   notch(center frequency, Q, dB isolation in stop-band)  
   ellip("filter type",order,dB ripple,db isolation, start frequency, stop frequency)

The current design's Q is unnecessarily low, which means that at 331 Hz, there is a 1.15 [deg] phase impact. Too big, if we're correcting for 1% level changes in the optical gain. Both new designs (the higher Q notch and the elliptic band stop) have much less phase impact at the highest calibration line frequency (the elliptic has 0.09 [deg] phase loss, the high-Q notch has only 0.06 [deg] phase loss). However, the elliptic bandstop gets much greater isolation in a broader, 30 Hz band, than the high-Q notch.

Thus, we proposed the elliptic band stop.

We will upgrade the filter bank, make a new DARM loop model, and update the front-end EPICs representation of that model necessary for correct computation of time-dependent correction factors tomorrow morning during maintenance.